Implicit vs Explicit Field Level Inference

Explore the fundamental approaches to field-level inference in cosmology through this 18-minute conference lecture that examines how to directly model cosmic density fields rather than relying on compressed summary statistics. Learn about the two primary methodologies: explicit field-level inference, which jointly reconstructs cosmological parameters and initial conditions to create complete "digital twins" of the Universe, and implicit field-level inference, which focuses solely on constraining cosmological parameters without full reconstruction. Understand how explicit methods enable detailed reconstruction of dark matter distributions and cosmic formation histories but require computationally intensive sampling in billion-dimensional parameter spaces, while implicit approaches achieve similar constraining power by extracting optimal data summaries in much lower-dimensional spaces. Discover how recent advances in machine learning and generative modeling are making high-dimensional cosmological inference computationally feasible, and examine the critical trade-offs between complete physical reconstruction versus computational efficiency. Gain insights into current challenges including the development of robust neural validation frameworks and scalable sampling methods for realistic forward models in modern cosmological analysis.